fix(common): anchor a file group's first slice on a committed log (#19785) A MOR file group can end up with its latest file slice keyed on a base instant that never committed. For example, under NBCC with the bucket index the earliest delta commit on a file group fails (or is rolled back) while later delta commits on the same group succeed. Log files are attributed to a slice by completion time, so when the group is first built the failed instant's log arrives first and opens a slice keyed on that uncommitted instant; every later committed log is then attributed to the same slice. isFileSliceCommitted only checks whether the slice's base instant itself committed, so the whole slice -- including its committed log files -- is treated as uncommitted and dropped from the reader view, silently losing committed data. Fix this at file-group construction instead of weakening the visibility gate. When the group has no slice yet, the earliest completed log now establishes the initial slice, so the slice is anchored on a real committed instant. An earlier pending log then follows the normal pending-log rule and attaches to that committed slice, instead of creating its own uncommitted slice that would also hide every later committed log in the group. isFileSliceCommitted keeps its original single invariant, so all existing and future view APIs benefit without per-call trimming. addLogFiles now owns the sort and the re-anchoring for a batch of log files; addLogFile becomes private and the file-system view calls the batch API. Sorting is done once and reused by both the re-anchor scan and the attach loop. Tests: - TestHoodieFileGroup#testUncommittedFirstLogDoesNotAnchorCommittedLogs verifies an uncommitted earliest log does not anchor the slice; the first committed log becomes the base instant and the pending log attaches to it. - TestHoodieFileGroup#testUncommittedBaseFileSliceStaysHiddenDespiteCommittedLogs verifies a slice with an uncommitted base file stays hidden even when later committed logs land in the same file group. - TestHoodieTableFileSystemView#testUncommittedFirstLogUsesFirstCommittedLogAsBaseInstant verifies the file-system view anchors the raw slice on the first committed log and surfaces only the committed logs. Closes #19774
Apache Hudi is an open data lakehouse platform, built on a high-performance open table format to ingest, index, store, serve, transform and manage your data across multiple cloud data environments.
Hudi stores all data and metadata on cloud storage in open formats, providing the following features across different aspects.
Hudi supports different types of queries, on top of a single table.
Learn more about Hudi at https://hudi.apache.org
Prerequisites for building Apache Hudi:
# Checkout code and build git clone https://github.com/apache/hudi.git && cd hudi mvn clean package -DskipTests -Dspark3.5 -Dflink2.2 # Start command spark-3.5.0-bin-hadoop3/bin/spark-shell \ --jars `ls packaging/hudi-spark-bundle/target/hudi-spark3.5-bundle_2.12-*.*.*-SNAPSHOT.jar` \ --conf 'spark.serializer=org.apache.spark.serializer.KryoSerializer' \ --conf 'spark.sql.extensions=org.apache.spark.sql.hudi.HoodieSparkSessionExtension' \ --conf 'spark.sql.catalog.spark_catalog=org.apache.spark.sql.hudi.catalog.HoodieCatalog' \ --conf 'spark.kryo.registrator=org.apache.spark.HoodieSparkKryoRegistrar'
To build for integration tests that include hudi-integ-test-bundle, use -Dintegration-tests.
To build the Javadoc for all Java and Scala classes (project should be already compiled):
# Javadoc generated under target/site/apidocs mvn javadoc:aggregate -Pjavadocs
The default Spark 3.x version, corresponding to spark3 profile is 3.5.3. The default Scala version is 2.12. Scala 2.13 is supported for Spark 3.5 and above.
Refer to the table below for building with different Spark and Scala versions.
| Maven build options | Expected Spark bundle jar name | Notes |
|---|---|---|
| (empty) | hudi-spark3.5-bundle_2.12 | For Spark 3.5.x and Scala 2.12 (default options) |
-Dspark3.3 | hudi-spark3.3-bundle_2.12 | For Spark 3.3.2+ and Scala 2.12 |
-Dspark3.4 | hudi-spark3.4-bundle_2.12 | For Spark 3.4.x and Scala 2.12 |
-Dspark3.5 -Dscala-2.12 | hudi-spark3.5-bundle_2.12 | For Spark 3.5.x and Scala 2.12 (same as default) |
-Dspark3.5 -Dscala-2.13 | hudi-spark3.5-bundle_2.13 | For Spark 3.5.x and Scala 2.13 |
-Dspark4.0 | hudi-spark4.0-bundle_2.13 | For Spark 4.0 and Scala 2.13 (Needs java 17) |
-Dspark4.1 | hudi-spark4.1-bundle_2.13 | For Spark 4.1 and Scala 2.13 (Needs java 17) |
-Dspark4.2 | hudi-spark4.2-bundle_2.13 | For Spark 4.2 and Scala 2.13 (Needs java 17) |
-Dspark3 | hudi-spark3-bundle_2.12 (legacy bundle name) | For Spark 3.5.x and Scala 2.12 |
Hudi uses ZSTD as the default Parquet compression codec with Spark 3.5 and newer. Spark 3.3 and 3.4 retain GZIP because Hudi's non-vectorized file-group reader uses parquet-java 1.12.x and can leak off-heap memory when reading ZSTD files (PARQUET-2160). Upgrade to Spark 3.5 or newer before using ZSTD. Keeping GZIP as the write default does not remove the risk when an older Spark runtime reads ZSTD files produced by another engine.
Please note that only Spark-related bundles, i.e., hudi-spark-bundle, hudi-utilities-bundle, hudi-utilities-slim-bundle, can be built using scala-2.13 profile. Hudi Flink bundle cannot be built using scala-2.13 profile. To build these bundles on Scala 2.13, use the following command:
# Build against Spark 3.5.x and Scala 2.13 mvn clean package -DskipTests -Dspark3.5 -Dscala-2.13 -pl packaging/hudi-spark-bundle,packaging/hudi-utilities-bundle,packaging/hudi-utilities-slim-bundle -am
For example,
# Build against Spark 3.5.x mvn clean package -DskipTests -Dspark3.5 -Dflink2.2 # Build against Spark 3.4.x mvn clean package -DskipTests -Dspark3.4 -Dflink2.2
Starting from versions 0.11, Hudi no longer requires spark-avro to be specified using --packages
The default Flink version supported is 2.2. The default Flink 2.2.x version, corresponding to the flink2.2 profile, is 2.2.1. Flink is Scala-free since 1.15.x, there is no need to specify the Scala version for Flink 1.15.x and above versions. Refer to the table below for building with different Flink and Scala versions.
| Maven build options | Expected Flink bundle jar name | Notes |
|---|---|---|
| (empty) | hudi-flink2.2-bundle | For Flink 2.2 (default options) |
-Dflink2.2 | hudi-flink2.2-bundle | For Flink 2.2 (same as default) |
-Dflink2.1 | hudi-flink2.1-bundle | For Flink 2.1 |
-Dflink2.0 | hudi-flink2.0-bundle | For Flink 2.0 |
-Dflink1.20 | hudi-flink1.20-bundle | For Flink 1.20 |
-Dflink1.19 | hudi-flink1.19-bundle | For Flink 1.19 |
-Dflink1.18 | hudi-flink1.18-bundle | For Flink 1.18 |
For example,
# Build against Flink 2.2.x mvn clean package -DskipTests -Dflink2.2
Unit tests can be run with maven profile unit-tests.
mvn -Punit-tests test
Functional tests, which are tagged with @Tag("functional"), can be run with maven profile functional-tests.
mvn -Pfunctional-tests test
Integration tests can be run with maven profile integration-tests.
mvn -Pintegration-tests verify
To run tests with spark event logging enabled, define the Spark event log directory. This allows visualizing test DAG and stages using Spark History Server UI.
mvn -Punit-tests test -DSPARK_EVLOG_DIR=/path/for/spark/event/log
Please visit https://hudi.apache.org/docs/quick-start-guide.html to quickly explore Hudi's capabilities using spark-shell.
Please check out our contribution guide to learn more about how to contribute. For code contributions, please refer to the developer setup.